IGedgeMAE: Topology-Aware Graph Masked Autoencoder with Dynamic Edge Importance Guidance

In the context of the digital integration of culture and tourism in Beiting, most of the existing related data mining technical methods only focus on the surface of the data, making it difficult to achieve high-quality mining effects and expressive capabilities. This seriously affects the construction of the digital platform. Therefore, in this paper, we study the potential of using graph structures as a masking strategy during the pre-training process to improve the model's ability to learn and mine high-quality data. Specifically, we introduce a new structural masking strategy, Importance-Guided Edge-based Masked Autoencoder, which aims to improve existing graph data mining models. It consists of two steps:(1) Node-based scoring: Evaluate each node in the graph and assign a score that reflects its structural importance. Then standardize this score as the final score of the node. The score of the edge connecting two nodes is the sum of the scores of the source node and the destination node.(2) Edge importance-guided masking: Utilizing the obtained edge importance scores, we propose a dynamic masking strategy from easy to difficult, which increases the awareness of structural learning in the self-supervised reconstruction task. In addition, we also introduce the process of reconstructing the node degree during the training process, which effectively guides the model to learn the graph structure information. Experimental data shows that, compared with traditional technical methods, the scheme based on IGedgeMAE performs better in multiple aspects, effectively promoting the digital integration development of the culture and tourism industry in the Beiting region.